Artificial Intelligence-RAN is moving from controlled lab environments into live networks, with Nvidia, Nokia and major operators treating a software-defined architecture as the path to future Artificial Intelligence-native wireless systems. Ahead of Mobile World Congress, running March 2-5 in Barcelona, Nvidia and Nokia announced new Artificial Intelligence-RAN collaborations with operators across Europe, Asia and North America, with T-Mobile U.S., SoftBank and Indosat Ooredoo Hutchison reaching field implementation milestones. New benchmarking from partners such as SynaXG indicates that Artificial Intelligence-RAN on Nvidia platforms can deliver carrier-grade reliability across multiple 5G spectrum bands, and over 20 Artificial Intelligence-RAN Alliance demos built on Nvidia technology are set to highlight performance gains, efficiency improvements and new edge Artificial Intelligence applications.
Operators are beginning to prove concurrent Artificial Intelligence and radio access workloads on shared infrastructure. T-Mobile U.S. demonstrated concurrent Artificial Intelligence and RAN processing on a Nvidia Artificial Intelligence-RAN platform using Nokia’s CUDA-accelerated software, with a 3.7GHz AirScale massive MIMO radio supporting commercial devices running video streaming, generative Artificial Intelligence and Artificial Intelligence-powered video captioning alongside 5G. SoftBank’s AITRAS live field trial achieved an industry-first, 16-layer massive MIMO using fully software-defined 5G on Nvidia’s Artificial Intelligence-RAN platform, while IOH progressed from proof of concept to pre-commercial validation using Nokia vRAN on Nvidia hardware, including Southeast Asia’s first Artificial Intelligence-powered 5G call enabling secure, real-time cross-border connectivity and remote robotic control. SynaXG demonstrated fully software-defined Artificial Intelligence-RAN with Nvidia Artificial Intelligence Aerial on a single Nvidia GH200 server across 4G, 5G sub-6GHz [FR1] and millimeter wave [FR2] bands, activating 20 component carriers with CU and DU on one platform, achieving a throughput of 36 Gbps and under 10 milliseconds latency.
The pace of Artificial Intelligence-RAN experimentation is accelerating, with this year’s Mobile World Congress featuring triple the number of Artificial Intelligence-RAN innovations over last year, and 26 out of 33 Artificial Intelligence-RAN Alliance demos built on Nvidia Artificial Intelligence Aerial and software-defined designs. Demonstrations span new neural air interfaces from DeepSig that show up to about 2x higher throughput and improved spectral and energy efficiency from the same spectrum, split-inferencing for robots and autonomous vehicles that allocates Artificial Intelligence tasks across device, edge and cloud, multi-tenant orchestration with zTouch Networks using Nvidia Multi-Instance GPU to share GPUs between Artificial Intelligence and RAN workloads, and Northeastern University and SoftBank’s Artificial Intelligence switching that flips in microseconds between Artificial Intelligence and classic algorithms. A broader ecosystem is forming around Nvidia Aerial RAN Computer platforms powered by Nvidia Grace CPUs and GPUs, with QCT, Supermicro, WNC, Eridan and LITEON introducing radios, servers and integrated solutions targeting 5G and 6G use cases. Nvidia’s latest State of Artificial Intelligence in Telecom report found that 77% of respondents anticipate a much faster time to deployment of this new Artificial Intelligence-native wireless architecture, and the company has open sourced its Aerial CUDA-accelerated RAN libraries and joined the OCUDU Ecosystem Foundation to advance open, secure, Artificial Intelligence-native 6G systems.
